fix(graphql): make type-filtered objects queries use sane plans

[tomxey]

Description of change

objects(filter: { type }) timed out on testnet for types with many objects (e.g. 0x3::staking_pool::StakedIota, 53k objects). Two fixes for this:

• Rewrite the changed-objects exclusion from LEFT JOIN (SELECT DISTINCT ...) to NOT EXISTS, so Postgres can check each row with one index lookup instead of re-scanning all changed objects per row (same change in the balance query). - This helps in case postgres picks a bad plan, where it thinks that there are not much candidate rows, where in reality there is way more of them.
• Add a migration raising the extended-statistics target on the object-type columns to 2500: testnet has more object types than fit in the current statistics, so rarer types got a rows=1 estimate and Postgres picked the bad plan. - This prevents postgres from picking the bad plan altogether

Links to any relevant issues

fixes #12536

How the change has been tested

• Basic tests (linting, compilation, formatting, unit/integration tests)
• Patch-specific tests (correctness, functionality coverage)

Measured on a staging server with a full testnet copy (921M rows in checkpointed_objects), across cursors of increasing age (= more objects changed after the viewed checkpoint), query objects(first: 3, filter: { type: "0x3::staking_pool::StakedIota" }):

cursor age(in terms of backward_history rows accumulated after it)	no fix	NOT EXISTS only	statistics only	both
~1k	3.7s	0.36s	43ms	43ms
~10k	24s	0.36s	53ms	44ms
30k–300k	timeout (>30s)	0.36–0.39s	73–329ms	40–78ms
~1M	1.2s	0.41s	1.1s	101ms
~8M (cursor ~3h old)	24s	8.3s	8.6s	7.9s

The planner estimate for the StakedIota filter went from 1 row to 40,627 (actual 52,458).
The migration took 1m41s on a testnet-size table.

Infrastructure QA (only required for crates that are maintained by @iotaledger/infrastructure)

• Synchronization of the indexer from genesis for a network including migration objects.
• Restart of indexer synchronization locally without resetting the database.
• Restart of indexer synchronization on a production-like database.
• Deployment of services using Docker.
• Verification of API backward compatibility.

Release Notes

• GraphQL: objects(filter: { type }) no longer times out for types with many objects.